Proposes a geometric-based debiasing algorithm for word embeddings, significantly reducing gender bias in Google News vectors while preserving semantic relations.

Tolga Bolukbasi, Kai-Wei Chang, James Zou et al.

Enriches skip-gram with character n-grams, improving multilingual word embeddings, especially for morphologically rich languages, with strong OOV handling.

Piotr Bojanowski, Edouard Grave, Armand Joulin et al.

fastText employs bag-of-words and hierarchical softmax, enabling training on billion-word datasets in under ten minutes with accuracy comparable to deep models.

Armand Joulin, Edouard Grave, Piotr Bojanowski et al.

Sequence-level knowledge distillation compresses large NMT models by 13×, speeds up decoding 10×, with only 0.4 BLEU loss, enabling fast greedy decoding.

Yoon Kim, Alexander M. Rush

Using Lempel-Ziv and NSB estimators on parallel corpora, the study identifies stable word entropy convergence points across 21 languages, enabling large-scale cross-linguistic entropy estimation.

Christian Bentz, Dimitrios Alikaniotis

Introduced LAMBADA dataset to evaluate broad discourse understanding; models score below 1%, highlighting significant gaps in long-range context comprehension.

Denis Paperno, Germán Kruszewski, Angeliki Lazaridou et al.

Introduces SQuAD, a large-scale dataset with 107,785 QA pairs, using dependency and syntactic tree features, achieving 51% F1; human performance is 86.8%.

Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev et al.

Introduces beam search optimization for seq2seq, addressing exposure and label bias, improving tasks like word ordering, parsing, and translation.

Sam Wiseman, Alexander M. Rush

Analysis of CNN/Daily Mail RC task shows simple models nearly reach performance ceiling, data noise limits further gains.

Danqi Chen, Jason Bolton, Christopher D. Manning

Learning language through SHRDLURN game interaction, model accuracy improved by 8%.

Sida I. Wang, Percy Liang, Christopher D. Manning

Using deep reinforcement learning for dialogue generation, improving coherence and informativeness.

Jiwei Li, Will Monroe, Alan Ritter et al.

On Yelp restaurant reviews, TF-IDF unigrams+bigrams with logistic regression reached 64% validation accuracy; test accuracy was 54% with RMSE 0.92.

Nabiha Asghar

Introduces the 'Story Cloze Test' framework using ROCStories corpus, highlighting causal understanding; baseline models perform poorly (~60%), deep models reach 70%.

Nasrin Mostafazadeh, Nathanael Chambers, Xiaodong He et al.

Introduces bidirectional LSTM-CRF and transition-based models for NER, achieving state-of-the-art results without language-specific resources.

Guillaume Lample, Miguel Ballesteros, Sandeep Subramanian et al.

Abstractive text summarization using sequence-to-sequence RNNs achieves state-of-the-art performance on two datasets.

Ramesh Nallapati, Bowen Zhou, Cicero Nogueira dos santos et al.

Introduces multiCluster and multiCCA methods for training multilingual embeddings across 59 languages, using dictionaries and monolingual data without parallel corpora.

Waleed Ammar, George Mulcaire, Yulia Tsvetkov et al.

Proposes Minimum Risk Training (MRT) for end-to-end neural machine translation, outperforming maximum likelihood estimation with up to +8.61 BLEU points.

Shiqi Shen, Yong Cheng, Zhongjun He et al.

Proposes PPDB-based universal paraphrastic sentence embeddings; simple models outperform LSTMs in cross-domain tasks.

John Wieting, Mohit Bansal, Kevin Gimpel et al.

Introducing BPE-based subword units for NMT, significantly improving rare word translation with 1.1-1.3 BLEU gains.

Rico Sennrich, Barry Haddow, Alexandra Birch

Character CNN combined with highway networks and LSTM achieves comparable performance to state-of-the-art with 60% fewer parameters, excelling across multiple languages.

Yoon Kim, Yacine Jernite, David Sontag et al.